View
4
Download
0
Category
Preview:
Citation preview
OutlineWhat makes a schedule serializable?Conflict serializabilityPrecedence graphsEnforcing serializability via 2-phase locking» Shared and exclusive locks» Lock tables and multi-level locking
Optimistic concurrency with validationConcurrency control + recoveryBeyond serializabilityCS 245 2
Example: Tj Ti
wj(A)ri(A)
Commit Ti
Abort Tj
Concurrency Control & Recovery
……
… ……
…
CS 245 3
Non-persistent commit (bad!)avoided byrecoverableschedules
Example: Tj Ti
wj(A)ri(A)wi(B)
Abort Tj[Commit Ti]
……
…
……
…
CS 245 4
Concurrency Control & Recovery
Cascading rollback (bad!)avoided byavoids-cascading-rollback (ACR)schedules
Core Problem
Schedule is conflict serializable
Tj Ti
But not recoverable
CS 245 5
To Resolve This
Need to mark the “final” decision for each transaction in our schedules:» Commit decision: system guarantees
transaction will or has completed» Abort decision: system guarantees
transaction will or has been rolled back
CS 245 6
Model This as 2 New Actions:
ci = transaction Ti commits
ai = transaction Ti aborts
CS 245 7
......
......
Tj Ti
wj(A)ri(A)
ci ¬ can we commit here?
Back to Example
CS 245 8
DefinitionTi reads from Tj in S (Tj ÞS Ti) if:
1. wj(A) <S ri(A)
2. aj <S r(A) (<S: does not precede)
3. If wj(A) <S wk(A) <S ri(A) then ak <S ri(A)
CS 245 9
Definition
Schedule S is recoverable if
whenever Tj ÞS Ti and j ¹ i and ci Î S
then cj <S ci
CS 245 10
Notes
In all transactions, reads and writes must precede commits or abortsó If ci Î Ti, then ri(A) < ai, wi(A) < ai
ó If ai Î Ti, then ri(A) < ai, wi(A) < ai
Also, just one of ci, ai per transaction
CS 245 11
How to Achieve Recoverable Schedules?
CS 245 12
With 2PL, Hold Write Locks Until Commit (“Strict 2PL”)
Tj Ti
wj(A)
cj
uj(A)ri(A)
CS 245 13
......
......
......
With Validation, No Change!
Each transaction’s validation point is its commit point, and only write after
CS 245 14
DefinitionsS is recoverable if each transaction commits only after all transactions from which it read have committed
S avoids cascading rollback if each transaction may read only those values written by committed transactions
S is strict if each transaction may read and write only items previously written by committed transactions (≡ strict 2PL)CS 245 15
Relationship of Recoverable, ACR & Strict Schedules
Avoids cascading rollback
Recoverable
ACR
Strict
Serial
CS 245 16
ExamplesRecoverable:
w1(A) w1(B) w2(A) r2(B) c1 c2
Avoids Cascading Rollback:w1(A) w1(B) w2(A) c1 r2(B) c2
Strict:w1(A) w1(B) c1 w2(A) r2(B) c2
CS 245 17
Recoverability & Serializability
Every strict schedule is serializable
Proof: equivalent to serial schedule based on the order of commit points» Only read/write from previously committed
transactions
CS 245 18
Recoverability & Serializability
CS 245 19
CS 245
OutlineWhat makes a schedule serializable?Conflict serializabilityPrecedence graphsEnforcing serializability via 2-phase locking» Shared and exclusive locks» Lock tables and multi-level locking
Optimistic concurrency with validationConcurrency control + recoveryBeyond serializability
20
Weaker Isolation Levels
Dirty reads: Let transactions read values written by other uncommitted transactions» Equivalent to having long-duration write locks,
but no read locks
Read committed: Can only read values from committed transactions, but they may change» Equivalent to having long-duration write locks
(X) and short-duration read locks (S)
CS 245 21
Weaker Isolation Levels
Repeatable reads: Can only read values from committed transactions, and each value will be the same if read again» Equivalent to having long-duration read &
write locks (X/S) but not table locks for insert
Remaining problem: phantoms!
CS 245 22
Weaker Isolation Levels
Snapshot isolation: Each transaction sees a consistent snapshot of the whole DB (as if we saved all committed values when it began)» Often implemented with multi-version
concurrency control (MVCC)
Still has some anomalies! Example?
CS 245 23
Weaker Isolation Levels
Snapshot isolation: Each transaction sees a consistent snapshot of the whole DB (as if we saved all committed values when it began)» Often implemented with multi-version
concurrency control (MVCC)
Write skew anomaly: txns write different values» Constraint: A+B ≥ 0» T1: read A, B; if A+B ≥ 1, subtract 1 from A» T2: read A, B; if A+B ≥ 1, subtract 1 from B» Problem: what if we started with A=1, B=0?
CS 245 24
Interesting Fact
Oracle calls its snapshot isolation level “serializable”, and doesn’t have the real thing!
CS 245 25
Why Distribute Our DB?
Store the same data item on multiple nodes to survive node failures (replication)
Divide data items & work across nodes to increase scale, performance (partitioning)
Related reasons:» Maintenance without downtime» Elastic resource use (don’t pay when unused)
CS 245 27
Outline
Replication strategies
Partitioning strategies
Atomic commitment & 2PC
CAP
Avoiding coordination
Parallel query executionCS 245 28
Outline
Replication strategies
Partitioning strategies
Atomic commitment & 2PC
CAP
Avoiding coordination
Parallel query executionCS 245 29
Replication
General problems:» How to tolerate server failures?» How to tolerate network failures?
CS 245 30
CS 245 31
Replication
Store each data item on multiple nodes!
Question: how to read/write to them?
CS 245 32
Primary-Backup
Elect one node “primary”
Store other copies on “backup”
Send requests to primary, which then forwards operations or logs to backups
Backup coordination is either:» Synchronous (write to backups before acking)» Asynchronous (backups slightly stale)
CS 245 33
Quorum Replication
Read and write to intersecting sets of servers; no one “primary”
Common: majority quorum» More exotic ones exist, like grid quorums
Surprise: primary-backupis a quorum too! C1: Write
C2: ReadCS 245 34
What If We Don’t Have Intersection?
CS 245 35
What If We Don’t Have Intersection?Alternative: “eventual consistency”» If writes stop, eventually all replicas will
contain the same data» Basic idea: asynchronously broadcast all
writes to all replicas
When is this acceptable?
CS 245 36
How Many Replicas?
In general, to survive F fail-stop failures, we need F+1 replicas
Question: what if replicas fail arbitrarily? Adversarially?
CS 245 37
What To Do During Failures?
Cannot contact primary?
CS 245 38
What To Do During Failures?
Cannot contact primary?» Is the primary failed?» Or can we simply not contact it?
CS 245 39
What To Do During Failures?
Cannot contact majority?» Is the majority failed?» Or can we simply not contact it?
CS 245 40
Solution to Failures
Traditional DB: page the DBA
Distributed computing: use consensus» Several algorithms: Paxos, Raft» Today: many implementations• Apache Zookeeper, etcd, Consul
» Idea: keep a reliable, distributed shared record of who is “primary”
CS 245 41
Consensus in a Nutshell
Goal: distributed agreement» On one value or on a log of events
Participants broadcast votes [for each event]» If a majority of notes ever accept a vote v,
then they will eventually choose v» In the event of failures, retry that round» Randomization greatly helps!
Take CS 244B for more details
CS 245 42
What To Do During Failures?
Cannot contact majority?» Is the majority failed?» Or can we simply not contact it?
Consensus can provide an answer!» Although we may need to stall…» (more on that later)
CS 245 43
Replication Summary
Store each data item on multiple nodes!
Question: how to read/write to them?» Answers: primary-backup, quorums» Use consensus to agree on operations or on
system configuration
CS 245 44
Outline
Replication strategies
Partitioning strategies
Atomic commitment & 2PC
CAP
Avoiding coordination
Parallel query executionCS 245 45
Partitioning
General problem:» Databases are big!» What if we don’t want to store the whole
database on each server?
CS 245 46
Partitioning Basics
Split database into chunks called “partitions”» Typically partition by row» Can also partition by column (rare)
Place one or more partitions per server
CS 245 47
Partitioning Strategies
Hash keys to servers» Random assignment
Partition keys by range» Keys stored contiguously
What if servers fail (or we add servers)?» Rebalance partitions (use consensus!)
Pros/cons of hash vs range partitioning?CS 245 48
What About Distributed Transactions?Replication:» Must make sure replicas stay up to date» Need to reliably replicate the commit log!
(use consensus or primary/backup)
Partitioning:» Must make sure all partitions commit/abort» Need cross-partition concurrency control!
CS 245 49
Outline
Replication strategies
Partitioning strategies
Atomic commitment & 2PC
CAP
Avoiding coordination
Parallel query executionCS 245 50
Atomic Commitment
Informally: either all participants commit a transaction, or none do
“participants” = partitions involved in a giventransaction
CS 245 51
So, What’s Hard?
CS 245 52
So, What’s Hard?
All the problems of consensus…
…plus, if any node votes to abort, all must decide to abort» In consensus, simply need agreement on
“some” value
CS 245 53
Two-Phase Commit
Canonical protocol for atomic commitment (developed 1976-1978)
Basis for most fancier protocols
Widely used in practice
Use a transaction coordinator» Usually client – not always!
CS 245 54
Two Phase Commit (2PC)1. Transaction coordinator sends prepare
message to each participating node
2. Each participating node responds to coordinator with prepared or no
3. If coordinator receives all prepared:» Broadcast commit
4. If coordinator receives any no:» Broadcast abort
CS 245 55
Informal Example
CS 245 56
Matei
Alice Bob
Pizza t
onight?
Sure
PizzaSpot
Confirmed
Pizz
a to
nigh
t?
Sure
Con
firm
ed
Got a table for 3 tonight?
Yes we do
I’ll book it
Case 1: Commit
CS 245 57UW CSE545
UW CSE545
Case 2: Abort
2PC + Validation
Participants perform validation upon receipt of prepare message
Validation essentially blocks between prepare and commit message
CS 245 59
2PC + 2PL
Traditionally: run 2PC at commit time» i.e., perform locking as usual, then run 2PC
to have all participants agree that the transaction will commit
Under strict 2PL, run 2PC before unlocking the write locks
CS 245 60
2PC + Logging
Log records must be flushed to disk on each participant before it replies to prepare» The participant should log how it wants to
respond + data needed if it wants to commit
CS 245 61
2PC + Logging Example
CS 245 62
Coordinator
Participant 1
Participant 2
<T1, Obj1, …>
read, write, etc
<T1, Obj3, …>
<T1, Obj2, …>
<T1, Obj4, …>
← log records
2PC + Logging Example
CS 245 63
Coordinator
Participant 1
Participant 2
<T1, Obj1, …>prep
are
<T1, Obj3, …>
<T1, Obj2, …>
<T1, Obj4, …>
<T1, ready>
<T1, ready>
prepareready
ready ← log records
<T1, commit>
2PC + Logging Example
CS 245 64
Coordinator
Participant 1
Participant 2
<T1, Obj1, …>co
mmit
<T1, Obj3, …>
<T1, Obj2, …>
<T1, Obj4, …>
<T1, ready>
<T1, ready>
commitdone
done ← log records
<T1, commit>
<T1, commit>
<T1, commit>
Optimizations Galore
Participants can send prepared messages to each other:» Can commit without the client» Requires O(P2) messages
Piggyback transaction’s last command on prepare message
2PL: piggyback lock “unlock” commands on commit/abort message
CS 245 65
What Could Go Wrong?
Coordinator
Participant Participant Participant
PREPARE
CS 245 66
What Could Go Wrong?
Coordinator
Participant Participant Participant
PREPARED PREPARED What if we don’thear back?
CS 245 67
Case 1: Participant UnavailableWe don’t hear back from a participant
Coordinator can still decide to abort» Coordinator makes the final call!
Participant comes back online?» Will receive the abort message
CS 245 68
What Could Go Wrong?
Participant Participant Participant
PREPARE
CS 245 69
Coordinator
What Could Go Wrong?
Participant Participant Participant
PREPARED PREPARED PREPARED
Coordinator does not reply!
CS 245 70
Case 2: Coordinator UnavailableParticipants cannot make progress
But: can agree to elect a new coordinator, never listen to the old one (using consensus)» Old coordinator comes back? Overruled by
participants, who reject its messages
CS 245 71
What Could Go Wrong?
Coordinator
Participant Participant Participant
PREPARE
CS 245 72
What Could Go Wrong?
Participant Participant Participant
PREPARED PREPARED
Coordinator does not reply!
No contact withthirdparticipant!
CS 245 73
Case 3: Coordinator and Participant UnavailableWorst-case scenario:» Unavailable/unreachable participant voted to
prepare» Coordinator hears back all prepare,
broadcasts commit» Unavailable/unreachable participant commits
Rest of participants must wait!!!
CS 245 74
Other Applications of 2PC
The “participants” can be any entities with distinct failure modes; for example:» Add a new user to database and queue a
request to validate their email» Book a flight from SFO -> JFK on United and
a flight from JFK -> LON on British Airways» Check whether Bob is in town, cancel my
hotel room, and ask Bob to stay at his place
CS 245 75
Coordination is Bad News
Every atomic commitment protocol is blocking(i.e., may stall) in the presence of:» Asynchronous network behavior (e.g.,
unbounded delays)•Cannot distinguish between delay and failure
» Failing nodes• If nodes never failed, could just wait
Cool: actual theorem!
CS 245 76
Outline
Replication strategies
Partitioning strategies
Atomic commitment & 2PC
CAP
Avoiding coordination
Parallel processingCS 245 77
CS 245 78Eric Brewer
Asynchronous Network Model
Messages can be arbitrarily delayed
Can’t distinguish between delayed messages and failed nodes in a finite amount of time
CS 245 79
CAP Theorem
In an asynchronous network, a distributed database can either:» guarantee a response from any replica in a
finite amount of time (“availability”) OR» guarantee arbitrary “consistency”
criteria/constraints about data
but not both
CS 245 80
CAP Theorem
Choose either:» Consistency and “Partition Tolerance”» Availability and “Partition Tolerance”
Example consistency criteria:» Exactly one key can have value “Matei”
“CAP” is a reminder:» No free lunch for distributed systems
CS 245 81
Why CAP is Important
Pithy reminder: “consistency” (serializability, various integrity constraints) is expensive!» Costs us the ability to provide “always on”
operation (availability)» Requires expensive coordination
(synchronous communication) even when we don’t have failures
CS 245 83
Recommended